Case Study: Revenue forecasting for ecommerce with Dipesh Shah

Dipesh Shah is a marketing consultant who links technology, data, and analytics to help businesses grow.

He was one of the very first Forecast Forge customers so I wanted to talk to him about how he has been forecasting sessions and revenue for one of his ecommerce clients.

For where this business is and what they want to look at, Forecast Forge gives them everything they need
— Dipesh Shah

I won't name the client here because Dipesh (and me) have shared actual revenue numbers from them on Twitter. In exchange for less transparency about who they are I can be more transparent about their numbers.

The client is an ecommerce business with a heavy focus on their membership offering. When they send a promotional email to their membership you definitely know about it!

Every client, every business model is going to have a slightly different way of understanding what they want and how they want it
— Dipesh Shah

Dipesh set up a regressor column with the historical dates of when marketing emails had been sent; these were easy to identify from the revenue figures. He also filled in the future dates for Q4 from the company's marketing calendar.

Using this setup he was able to quickly make a forecast for the daily sessions in Q4.

The revenue forecast using this method looked unresonably low. Dipesh used the same regressor columns but added a log transform after which the revenue forecast looked much more sensible.

The predicted revenue for Q4 was £262,299.

The actual revenue for Q4 was £291,788. An error of 10.1%.

After reviewing the forecast against actual performance Dipesh has been working on several improvements to forecast Q1 2021 and beyond.

If you're updating a forecast for a year, you can probably do it within an hour or so
— Dipesh Shah

The first thing he noticed was that some promotional emails generated a lot more revenue than others. These promotions can be identified in advance which will make the predictions a lot more accurate for those days.

The client switched ecommerce platform in early November 2020. After this change their daily traffic and revenue had spikes that were a lot higher than what was happening before the replatform. Because the machine learning model had never seen this data before it was not part of its prediction; but this data is included for the forecasts for 2021 so the predictions will be better.

Forecast Forge gives them everything they want at a low cost to advance their analytics further down the line
— Dipesh Shah